Return to MUSA 801 Projects Page
This project was produced as part of the University of Pennsylvania’s Master of Urban Spatial Analytics Spring 2019 Practicum (MUSA 801) taught by Ken Steif, Michael Fichman, and Matt Harris. We would like to thank the Philadelphia Fire Department for providing useful information and data.
This document is intended to enable others to replicate the methodology of a study of structural fires and hydrant inspections. We first introduce the context of this project, followed by the data, methodology, modeling, and guiding appendices. A table of contents is provided for easy navigation, along with select hyperlinks throughout the document.
According to the data from Philadelphia fire department, the number of fires in Philadelphia is increasing year by year since 2015. How to inspect fire hydrants more efficiently and deal with fire risk has become a new challenge for the fire department. Besides,the spike in the number of fires after 2017 was also a result of the change in the procedure for recording fires. More information can be found in this article.
A hydrant that is more likely in need of inspection could be the last on the list, and firefighters have to take time and energy to inspect each hydrant. To address this, we propose using past data on fires and risk factors of fires to identify areas in Philadelphia where fires are more likely. Moreover, based on our modelling result, we also include about using social, population to build environment factors to extend the fire model to a prioritization to identify which hydrants are more urgent to be maintained. Taking into consideration both fire risk and hydrant maintenance, we aim to generate a list for local engines to inspect hydrants and for the PWD to then maintain them. This would ensure that most hydrants are in good condition most of the time and allows firefighters to focus on a few hydrants rather than inspecting all hydrants.
To summarize, the use case is:
Prioritize inspections for local fire engines
Ensure that at all times, especially in areas at high risk of fire, most hydrants are in good condition
Improve the inspection efficiency and reduce the need to inspect good hydrants
Our goal is to provide decision-makers with insight regarding the built-environment and intrinsic characteristics that contribute to the priority of fire hydrant inspections with the distribution of fire fire risk over the City. Therefore, besides fire and hydrant information, factors that may outline the fire causes are considered in our data selection, including public safety, demographics, properties, facilities and environment. For our analysis, we primarily use:
Data from Philadelphia Fire Department and Philadelphia Water Department in order to work on the most accurate and updated information for fire incidents and fire hydrants;
Open data source (OpenDataPhilly) so that other municipalities with similar open data repositories can refer to or even reproduce our analysis.
The data was also “wrangled” before being explored in the following section. This process included various transformations of the data in order to optimize predictive ability of each variable. For details on this procedure, please see Feature Engineering part. Through exploratory analysis and the modeling process, the final dataset was narrowed down to a set of best predictors. For results on how the data ultimately performed in the models, please see Model Building.
A hydrant should have a higher priority of being inspected if
there is a latent risk of fire in the vicinity;
the hydrant is close to social impact factors such as the proximity of vulnerable population and certain facilities, as it is not feasible to model hydrant conditions.
By considering different aspects, we decided to use features for the year 2018 to build the model, and test it by using the same features for the year 2019.
To best extrapolate the relationship between features and fires, we “engineered” features to include in our fishnet. Many of these variables have more explanatory power as varied spatial measurements, by aggregating and engineering features, we adopted:
1. Kernel Density: This is the average spatial density of the certain features per grid cell.
2. Count: This is the count of certain features per grid cell.
3. Nearest Neighbor Distance: The average distance from each grid cell to the nearest certain number of features.
The features used in the model were classified into four categories: built environment factor, risk factor, demographic factor and time-spatial factor.For an explanation of which measurements were used for each variable, please see Appendix: Data Dictionary.
In 1803 Frederick Graff, Sr., designed for the then recently constructed Philadelphia hydrant, a stand-pipe intended to remain permanently in position and to be constantly charged with water.After that, the hydrants in Philadelphia continued to expand as the city grew:In the 1900s, fire hydrants were mainly located in the city center; In the 1950s, the distribution of fire hydrants gradually began to spread around the city; In the 2000s, the fire hydrants had basically covered the whole city, and we can find that the fire hydrants density in the downtown area is greater than the surrounding area.
In the present day, there is the highest density of hydrants in center city and the density seems to dissipate outwards from the center city. This could mean that fire fighters spend more time inspecting hydrants in the center city due to the higher concentration. Of course, the high fire hydrant density in central city may also be related to the high frequency of fire accident in the downtown area.